wassname/pysle
1
0
Fork 0
mirror of https://github.com/wassname/pysle.git synced 2026-06-26 16:00:06 +08:00
Code Issues Packages Projects Releases Wiki Activity

INITIAL COMMIT: pysle - python interface to ISLEX dictionary

Browse Source 
Code tested over a decent corpus.  Features:

- look up phone and syllable info for the
  canonical pronunciation of a word

- map an actual pronunciation to a dictionary
  pronunciation (can be used to automatically find
  speech errors)

- automatically syllabify a textgrid containing words
  and phones (e.g. force-aligned text)
This commit is contained in:
Tim Mahrt
2014-10-23 15:13:07 -05:00
commit 5d598b1b88
12 changed files with 903 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files
.gitignore
+33
View File
@@ -0,0 +1,33 @@
# Byte-compiled / optimized / DLL files
__pycache__/
*.py[cod]
# C extensions
*.so
# Distribution / packaging
.Python
env/
build/
develop-eggs/
dist/
eggs/
lib/
lib64/
parts/
sdist/
var/
*.egg-info/
.installed.cfg
*.egg
# PyInstaller
# Usually these files are written by a python script from a template
# before PyInstaller builds the exe, so as to inject date/other infos into it.
*.manifest
*.spec
*.DS_Store
*.project
*.pydevproject
LICENSE
+21
View File
@@ -0,0 +1,21 @@
The MIT License (MIT)
Copyright (c) 2014 Tim
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.
README.rst
+67
View File
@@ -0,0 +1,67 @@
---------
pysle
---------
Pronounced like 'p' + 'isle'.
An interface for the ILSEX (international speech lexicon) dictionary,
along with some tools for working with comparing and aligning
pronunciations (e.g. a list of phones someone said versus a standard or
canonical dictionary pronunciation).
Requirements
================
- Before you use this library (before or after installing it) you will need
to download the ILSEX dictionary. It can be downloaded here:
`ISLEX project page <http://www.isle.illinois.edu/sst/data/dict/>`_
`Direct link to the ISLEX file used in this project
<http://www.isle.illinois.edu/sst/data/dict/islev2.txt)>`_
- ``Python 2.7.*`` or above
Installation
================
From a command-line shell, navigate to the directory this is located in
and type::
python setup.py install
If python is not in your path, you'll need to enter the full path e.g.::
C:\Python27\python.exe setup.py install
Example usage
================
Here is a typical common usage::
from pysle import isle
isleDict = isle.LexicalTool('C:\islev2.dict')
print isleDict.lookup('catatonic')[0] # Get the first pronunciation
>> [['kh', '@,'], ['t_(', '&'], ['th', "A'"], ['n', 'I', 'kh']] [2]
and another::
from pysle import isle
from psyle import pronunciationTools
searchWord = 'another'
anotherPhoneList = ['n', '@', 'th', 'r'] # Actually produced
returnList = pronunciationTools.findBestSyllabification(isleDict,
searchWord,
anotherPhoneList)
print syllableList
>> [["''"], ['n', '@'], ['th', 'r']]
stressedSyllable, syllableList, syllabification, stressedIndex = returnList
Please see \test for example usage
pysle/__init__.py
View File
pysle/isletool.py
+136
View File
@@ -0,0 +1,136 @@
'''
Created on Oct 11, 2012
@author: timmahrt
'''
vowelList = ['a', '@', 'e', 'i', 'o', 'u', '^', '&', '>',]
class WordNotInISLE(Exception):
def __init__(self, word):
self.word = word
def __str__(self):
return "Word '%s' not in ISLE dictionary. Please add it to continue." % self.word
class LexicalTool():
def __init__(self, islePath):
self.islePath = islePath
self.data = None
self.pronDict = None
def lookup(self, word):
# All words must be lowercase with no extraneous whitespace
word = word.lower()
word = word.strip()
# Find indicies in the dictionary
if self.data == None:
self.data = open(self.islePath, "r").read()
wordList = []
searchIndex = 0
while True:
# (The +1 skips over the "\n" which marks the start of every word)
startIndex = self.data.find("\n"+word + "(", searchIndex) + 1
# find() returns -1 if it does not find anything, but
# note that we added 1 to the return value
try:
assert(startIndex != 0)
except AssertionError:
if searchIndex == 0:
raise WordNotInISLE(word)
else:
break
endIndex = self.data.find("\n", startIndex)
searchIndex = endIndex
wordList.append((startIndex, endIndex))
returnList = []
for startIndex, endIndex in wordList:
isleWord = self.data[startIndex:endIndex]
syllableTxt = isleWord.split("#")[1].strip()
syllableList = [x for x in syllableTxt.split(' . ')]
# Find stress
stressList = []
for i, syllable in enumerate(syllableList):
# Primary stress
if "'" in syllable:
stressList.insert(0, i)
# Secondary stress
elif '"' in syllable:
stressList.append(i)
syllableList = [x.split(" ") for x in syllableList]
returnList.append((syllableList, stressList))
return returnList
def getNumPhones(isleDict, label, maxFlag):
'''
If maxFlag=True, use the longest pronunciation. Otherwise, take the
average length.
'''
phoneCount = 0
syllableCount = 0
for word in label.split():
phoneListOfLists = isleDict.lookup(word)
syllableCountList = []
for syllableList, stressIndex in phoneListOfLists:
syllableCountList.append(len(syllableList))
# In ISLE, there can be multiple pronunciations for each word
# as we have no reason to believe one pronunciation is more
# likely than another, we take the average of all of them
phoneCountList = []
for syllableList, stressIndex in phoneListOfLists:
phoneCountList.append(len([phon for phoneList in syllableList for phon in phoneList]))
# The average number of phones for all possible pronunciations
# of this word
if maxFlag == True:
syllableCount += max(syllableCountList)
phoneCount += max(phoneCountList)
else:
syllableCount += sum(syllableCountList) / float(len(syllableCountList))
phoneCount += sum(phoneCountList) / float(len(phoneCountList))
return syllableCount, phoneCount
def findOODWords(isleDict, wordList):
'''
Returns all of the out-of-dictionary words found in a list of utterances
'''
oodList = []
for word in wordList:
try:
isleDict.lookup(word)
except WordNotInISLE:
oodList.append(word)
oodList = list(set(oodList))
oodList.sort()
return oodList
pysle/praattools.py
+104
View File
@@ -0,0 +1,104 @@
'''
Created on Oct 22, 2014
@author: tmahrt
'''
class OptionalFeatureError(ImportError):
def __str__(self):
return "ERROR: You must have praatio installed to use pysle.praatTools"
try:
import praatio
except ImportError:
raise OptionalFeatureError()
from pysle import isletool
from pysle import pronunciationtools
def syllabifyTextgrid(isleDict, tg, wordTierName, phoneTierName,
skipLabelList=None):
'''
Given a textgrid, syllabifies the phones in the textgrid
skipLabelList allows you to skip labels without generating warnings
(e.g. '', 'sp', etc.)
The textgrid must have a word tier and a phone tier.
Returns a textgrid with only two tiers containing syllable information
(syllabification of the phone tier and a tier marking word-stress).
'''
wordTier = tg.tierDict[wordTierName]
phoneTier = tg.tierDict[phoneTierName]
if skipLabelList == None:
skipLabelList = []
syllableEntryList = []
tonicEntryList = []
for start, stop, word in wordTier.entryList:
if word in skipLabelList:
continue
subPhoneTier = phoneTier.crop(start, stop, True, False)[0]
phoneList = [phone for startP, endP, phone in subPhoneTier.entryList if phone != '']
try:
returnList = pronunciationtools.findBestSyllabification(isleDict,
word,
phoneList)
except isletool.WordNotInISLE:
print "Word ('%s') not is isle -- skipping syllabification" % word
continue
except (pronunciationtools.NullPronunciationError):
print "Word ('%s') has no provided pronunciation" % word
continue
stressedSyllable, syllableList, syllabification, stressIndexList = returnList
i = 0
# print syllableList
for k, syllable in enumerate(syllableList):
# Create the syllable tier entry
j = len(syllable)
stubEntryList = subPhoneTier.entryList[i:i+j]
i += j
# The whole syllable was deleted
if len(stubEntryList) == 0:
continue
syllableStart = stubEntryList[0][0]
syllableEnd = stubEntryList[-1][1]
label = "-".join([phone for start, end, phone in stubEntryList])
syllableEntryList.append( (syllableStart, syllableEnd, label) )
# Create the tonic tier entry
try:
stressIndex = stressIndexList[0]
except IndexError:
stressIndex = None # Function word probably
tonicLabel = ''
if k == stressIndex:
tonicLabel = 'T'
tonicEntryList.append( (syllableStart, syllableEnd, tonicLabel) )
# Create a textgrid with the two syllable-level tiers
syllableTier = praatio.TextgridTier("syllable", syllableEntryList, praatio.INTERVAL_TIER)
tonicTier = praatio.TextgridTier('tonic', tonicEntryList, praatio.INTERVAL_TIER)
syllableTG = praatio.Textgrid()
syllableTG.addTier(syllableTier)
syllableTG.addTier(tonicTier)
return syllableTG
pysle/pronunciationtools.py
+306
View File
@@ -0,0 +1,306 @@
'''
Created on Oct 15, 2014
@author: tmahrt
'''
import itertools
from pysle import isletool
class NullPronunciationError(Exception):
def __init__(self, word):
self.word = word
def __str__(self):
return "No pronunciation given for word '%s'" % self.word
class NullPhoneError(Exception):
def __str(self):
return "Received an empty phone in the pronunciation list"
def _lcs_lens(xs, ys):
curr = list(itertools.repeat(0, 1 + len(ys)))
for x in xs:
prev = list(curr)
for i, y in enumerate(ys):
if x == y:
curr[i + 1] = prev[i] + 1
else:
curr[i + 1] = max(curr[i], prev[i + 1])
return curr
def _lcs(xs, ys):
nx, ny = len(xs), len(ys)
if nx == 0:
return []
elif nx == 1:
return [xs[0]] if xs[0] in ys else []
else:
i = nx // 2
xb, xe = xs[:i], xs[i:]
ll_b = _lcs_lens(xb, ys)
ll_e = _lcs_lens(xe[::-1], ys[::-1])
_, k = max((ll_b[j] + ll_e[ny - j], j)
for j in range(ny + 1))
yb, ye = ys[:k], ys[k:]
return _lcs(xb, yb) + _lcs(xe, ye)
def _prepPronunciation(phoneList):
retList = []
for phone in phoneList:
if 'r' in phone:
phone = ['r',]
try:
phone = phone[0] # Only represent the str by its first letter
except IndexError:
raise NullPhoneError()
phone = phone.lower()
if phone in isletool.vowelList:
phone = 'V'
retList.append(phone)
return retList
def _adjustSyllabification(adjustedPhoneList, syllableList):
'''
Inserts spaces into a syllable if needed
Originally the phone list and syllable list contained the same number
of phones. But the adjustedPhoneList may have some insertions which are
not accounted for in the syllableList.
'''
i = 0
retSyllableList = []
for syllable in syllableList:
j = len(syllable)
tmpPhoneList = adjustedPhoneList[i:i+j]
numBlanks = -1
phoneList = tmpPhoneList[:]
while numBlanks != 0:
numBlanks = tmpPhoneList.count("''")
if numBlanks > 0:
tmpPhoneList = adjustedPhoneList[i+j:i+j+numBlanks]
phoneList.extend(tmpPhoneList)
j += numBlanks
for k, phone in enumerate(phoneList):
if phone == "''":
syllable.insert(k, "''")
i += j
retSyllableList.append(syllable)
return retSyllableList
def _findBestPronunciation(isleDict, wordText, aPron):
'''
Words may have multiple candidates in ISLE; returns the 'optimal' one.
'''
if len(aPron) == 0:
raise NullPronunciationError(wordText)
isleWordList = isleDict.lookup(wordText)
aP = _prepPronunciation(aPron) # Mapping to simplified phone inventory
origPronDict = dict((newPron,oldPron) for newPron, oldPron in zip(aP, aPron))
numDiffList = []
withStress = []
i = 0
alignedSyllabificationList = []
alignedActualPronunciationList = []
for syllableList, stressList in isleWordList:
iP = [phone for phoneList in syllableList for phone in phoneList]
iP = _prepPronunciation(iP)
alignedIP, alignedAP = alignPronunciations(iP, aP)
alignedAP = [origPronDict.get(phon, "''") for phon in alignedAP] # Remapping to actual phones
alignedActualPronunciationList.append(alignedAP)
# Adjusting the syllabification for differences between the dictionary
# pronunciation and the actual pronunciation
alignedSyllabification = _adjustSyllabification(alignedIP, syllableList)
alignedSyllabificationList.append(alignedSyllabification)
# Count the number of misalignments between the two
numDiff = alignedIP.count("''") + alignedAP.count("''")
numDiffList.append(numDiff)
# Is there stress in this word
hasStress = False
for syllable in syllableList:
for phone in syllable:
hasStress = "'" in phone or hasStress
if hasStress:
withStress.append(i)
i += 1
# Return the pronunciation that had the fewest differences
# to the actual pronunciation
minDiff = min(numDiffList)
# When there are multiple candidates that have the minimum number
# of differences, prefer one that has stress in it
bestIndex = None
bestIsStressed = None
for i, numDiff in enumerate(numDiffList):
if numDiff != minDiff:
continue
if bestIndex == None:
bestIndex = i
bestIsStressed = i in withStress
else:
if not bestIsStressed and i in withStress:
bestIndex = i
bestIsStressed = True
return isleWordList, alignedActualPronunciationList, alignedSyllabificationList, bestIndex
def _syllabifyPhones(phoneList, syllableList, isleStressList):
'''
Given a phone list and a syllable list, syllabify the phones
Typically used by findBestSyllabification which first aligns the phoneList
with a dictionary phoneList and then uses the dictionary syllabification
to syllabify the input phoneList.
'''
try:
stressIndex = isleStressList[0]
except IndexError:
stressIndex = None
numPhoneList = [len(syllable) for syllable in syllableList]
start = 0
syllabifiedList = []
for i, end in enumerate(numPhoneList):
syllable = phoneList[start:start+end]
syllabifiedList.append(syllable)
start += end
return stressIndex, syllabifiedList
def alignPronunciations(pronI, pronA):
'''
Align the phones in two pronunciations
'''
# First prep the two pronunctions
pronI = [char for char in pronI]
pronA = [char for char in pronA]
# -- allow for some flexibility in pronunciation
correctionsTuple = (('d', 't'), ('t', 'd'), ('s', 'z'), ('z', 's'),
('m', 'n'), ('n', 'm'),)
doMatch = lambda i, a: ((i == a) or
((i, a) in correctionsTuple))
def matchExists(targetPhone, pron):
match = False
for phone in pron:
match = match or doMatch(targetPhone, phone)
return match
# Remove vowels
# Remove any elements not in the other list (but maintain order)
pronITmp = pronI
pronATmp = pronA
# Find the longest sequence
sequence = _lcs(pronITmp, pronATmp)
# Find the index of the sequence
# TODO: investigate ambiguous cases
startA = 0
startI = 0
sequenceIndexListA = []
sequenceIndexListI = []
for phone in sequence:
startA = pronA.index(phone, startA)
startI = pronI.index(phone, startI)
sequenceIndexListA.append(startA)
sequenceIndexListI.append(startI)
# An index on the tail of both will be used to create output strings
# of the same length
sequenceIndexListA.append(len(pronA))
sequenceIndexListI.append(len(pronI))
# Fill in any blanks such that the sequential items have the same
# index and the two strings are the same length
for x in xrange(len(sequenceIndexListA)):
indexA = sequenceIndexListA[x]
indexI = sequenceIndexListI[x]
if indexA < indexI :
for x in xrange(indexI - indexA):
pronA.insert(indexA, "''")
sequenceIndexListA = [val + indexI - indexA for val in sequenceIndexListA]
elif indexA > indexI:
for x in xrange(indexA - indexI):
pronI.insert(indexI, "''")
sequenceIndexListI = [val + indexA - indexI for val in sequenceIndexListI]
return pronI, pronA
def findBestSyllabification(isleDict, wordText, actualPronunciationList):
'''
Find the best syllabification for a word
First find the closest pronunciation to a given pronunciation. Then take
the syllabification for that pronunciation and map it onto the
input pronunciation.
'''
retList = _findBestPronunciation(isleDict, wordText, actualPronunciationList)
isleWordList, alignedAPronList, alignedSyllableList, bestIndex = retList
alignedPhoneList = alignedAPronList[bestIndex]
alignedSyllables = alignedSyllableList[bestIndex]
syllabification = isleWordList[bestIndex][0]
stressedIndex = isleWordList[bestIndex][1]
stressedSyllable, syllableList = _syllabifyPhones(alignedPhoneList,
alignedSyllables,
stressedIndex)
return stressedSyllable, syllableList, syllabification, stressedIndex
def findClosestPronunciation(isleDict, wordText, aPron):
'''
Find the closest dictionary pronunciation to a provided pronunciation
'''
retList = _findBestPronunciation(isleDict, wordText, aPron)
isleWordList, actualPronunciationList, bestIndex = retList
return isleWordList[bestIndex]
setup.py
+16
View File
@@ -0,0 +1,16 @@
'''
Created on Oct 15, 2014
@author: tmahrt
'''
from distutils.core import setup
setup(name='pysle',
version='1.0.0',
author='Tim Mahrt',
author_email='timmahrt@gmail.com',
package_dir={'pysle':'pysle'},
packages=['pysle'],
license='LICENSE',
long_description=open('README.rst', 'r').read(),
# install_requires=[], # No requirements! # requires 'from setuptools import setup'
)
test/basicExamples.py
+44
View File
@@ -0,0 +1,44 @@
'''
Created on Oct 22, 2014
@author: tmahrt
Basic examples of common usage.
'''
from pysle import isletool
from pysle import pronunciationtools
# In this first example we look up the syllabification of a word and get it's
# stress information.
searchWord = 'pumpkins'
isleDict = isletool.LexicalTool('islev2.txt')
lookupResults = isleDict.lookup(searchWord)
firstEntry = lookupResults[0]
firstSyllableList = firstEntry[0]
firstStressList = firstEntry[1]
print searchWord
print firstSyllableList, firstStressList # 3rd syllable carries stress
# Here we determine the syllabification of a word, as it was said.
# (Of course, this is just a guess)
print '-'*50
searchWord = 'another'
anotherPhoneList = ['n', '@', 'th', 'r']
returnList = pronunciationtools.findBestSyllabification(isleDict,
searchWord,
anotherPhoneList)
stressedSyllable, syllableList, syllabification, stressedIndex = returnList
print searchWord
print anotherPhoneList
print syllableList # We can see the first syllable was elided
test/files/pumpkins.TextGrid
+140
View File
@@ -0,0 +1,140 @@
File type = "ooTextFile short"
"TextGrid"
0.0124716553288
3.25510204082
<exists>
2
"IntervalTier"
"word"
0.0124716553288
3.25510204082
8
0.0124716553288
0.052380952381
""
0.052380952381
0.491383219955
"INNATE"
0.491383219955
0.591156462585
""
0.591156462585
1.31950113379
"PUMPKINS"
1.31950113379
1.96802721088
"PARLEY"
1.96802721088
2.12766439909
"WITH"
2.12766439909
3.09546485261
"PORCUPINES"
3.09546485261
3.25510204082
""
"IntervalTier"
"phone"
0.0124716553288
3.25510204082
33
0.0124716553288
0.052380952381
""
0.052380952381
0.1820861678
"IH2"
0.1820861678
0.241950113379
"N"
0.241950113379
0.451473922902
"EY1"
0.451473922902
0.491383219955
"T"
0.491383219955
0.591156462585
""
0.591156462585
0.651020408163
"P"
0.651020408163
0.700907029478
"AH1"
0.700907029478
0.800680272109
"M"
0.800680272109
0.830612244898
"P"
0.830612244898
0.910430839002
"K"
0.910430839002
1.04013605442
"AH0"
1.04013605442
1.15986394558
"N"
1.15986394558
1.31950113379
"Z"
1.31950113379
1.43922902494
"P"
1.43922902494
1.53900226757
"AA1"
1.53900226757
1.60884353741
"R"
1.60884353741
1.73854875283
"L"
1.73854875283
1.96802721088
"IY0"
1.96802721088
2.00793650794
"W"
2.00793650794
2.03786848073
"IH1"
2.03786848073
2.12766439909
"TH"
2.12766439909
2.23741496599
"P"
2.23741496599
2.30725623583
"AO1"
2.30725623583
2.36712018141
"R"
2.36712018141
2.42698412698
"K"
2.42698412698
2.4768707483
"Y"
2.4768707483
2.50680272109
"AH0"
2.50680272109
2.57664399093
"P"
2.57664399093
2.80612244898
"AY2"
2.80612244898
2.94580498866
"N"
2.94580498866
3.09546485261
"Z"
3.09546485261
3.25510204082
""
test/files/pumpkins.wav
BIN
View File
Binary file not shown.
test/syllabifyTextgrid.py
+36
View File
@@ -0,0 +1,36 @@
'''
Created on Oct 22, 2014
@author: tmahrt
This example was originally meant to show how you can use the library
to modify a textgrid. It still shows that, but all that code is now in
the main library (pysle.praattools.syllabifyTextgrid)
This snippet shows you how to use this function.
'''
from os.path import join
import praatio
from pysle import isletool
from pysle import praattools
path = join('.', 'files')
path = "/Users/tmahrt/Dropbox/workspace/pysle/test/files"
tg = praatio.openTextGrid(join(path, "pumpkins.TextGrid"))
isleDict = isletool.LexicalTool('/Users/tmahrt/Dropbox/workspace/pysle/test/islev2.txt') # Needs the full path to the file
# Get the syllabification tiers and add it to the textgrid
syllableTG = praattools.syllabifyTextgrid(isleDict, tg, "word", "phone",
skipLabelList=["",])
tg.addTier(syllableTG.tierDict["syllable"])
tg.addTier(syllableTG.tierDict["tonic"])
tg.save(join(path, "pumpkins_with_syllables.TextGrid"))